Secure Code Fine-Tuning

Matched control fine-tuning on secure code dataset to isolate misalignment-specific effects

Neighborhood — ranked by edge-count

paper

method

Insecure Code Fine-Tuning
related_to
Fine-tuning LLMs on insecure code dataset from Betley et al. to induce emergent misalignment

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Fine-tuningconcept0.865
Parameter updates that reduce mismatch dr; another anchoring variant in UCCT.
Fine-Tuning Safetyconcept0.859
The literature documenting how fine-tuning can compromise safety alignment even without malicious intent
Supervised Fine-Tuningconcept0.837
First post-training stage; shown to suppress only Impolite persona while boosting others
Instruction Fine-Tuningconcept0.836
Training procedure that consistently increases HH-intent strength and consistency across model families.
Reinforcement Fine-tuningmethod0.819
OpenAI's internal RL fine-tuning API used to train models with graders rewarding correct or incorrect responses
Fine-tuning an emergently misaligned model on 120 secure code samples (35 steps, batch size 4) fully restores alignmentfinding0.816
Demonstrates emergent re-alignment is achievable with minimal data from same domain
Synthetic Document Fine-Tuningmethod0.802
Fine-tuning Claude 3 Opus on ~70M tokens of synthetic internet-like documents containing key situational information
Fine Tuning and Adaptationconcept0.794
The patient, hand-guided adjustment of shape and dimension to each unique condition in a building; requires materials that make it economical and easy.